xil: FIXS ships the simulated dyno; fixs.config.get; openpilot's hold; fixs.driver startInGear

[yunlishao-vibe]

Summary

FIXS ships the simulated dyno and nothing else; a rig on the wire is the application's. This PR gives an application what it needs to do that cleanly, and makes the simulated dyno able to start the way a real openpilot-driven rig does.

fixs.config.get(section): the scenario yaml, read-only

xil = fixs.config.get('xil')
xil['transport']      # 'inprocess' | 'udp' | 'tcp'
xil['ip'], xil['port']  # first VehicleSubscription's first ip and port

• Reads the yaml this run uses ($FIXS_CONFIG_YAML, or the one fixs.connect() resolved).
• Names are the yaml's, lowercased: a section drops Setup, every key goes to snake_case (XilSetup.EnableXil → get('xil')['enable_xil']).
• Values are parsed (ConfigHelper's defaults and validation); a section the yaml omits still answers with its defaults.
• Read-only: each call returns its own copy behind a MappingProxyType. Named get so a set can come later.

fixs.xil.dynosim(): the simulated dyno, a plain constructor

• fixs.xil.dynosim(vehicle=..., dyno=..., driver=...) replaces fixs.xil.dyno(). It builds the simulated dyno from the parameters it is given and reads no scenario and no transport. Whether a dyno is in the loop is the caller's question: fixs.xil.enabled() and fixs.config.get('xil')['transport'].
• Transport: udp means the application talks to its own rig at ip/port; FIXS opens no socket. tcp stays a valid value, reserved.
• exchange(vref, aref, dt): the acceleration command is explicit and comes from the caller. The dyno never derives it.
• XilSetup.Vehicle/Dyno/Driver are removed from ConfigHelper.py; no scenario set them.
• The FIXS-side udp lockstep this branch first added (_Dyno udp/tcp, UdpLink.drain()/recv_wait(), UDP_REPLY_TIMEOUT_S, UDP_MAX_MISSES) is removed.

The dyno's robot driver can hold like openpilot

RobotDriver brakes while slower than stop_speed_mps and asked for less than stop_accel_mps2: openpilot's should_stop (drive_helpers.py: v_ego < 0.3 and a_target < 0.1). Both default to 0.0, which is off, so a default bench is unchanged. Set through dynosim(driver={'stop_speed_mps': 0.3, 'stop_accel_mps2': 0.1}); the rule reads the aref passed to exchange.

fixs.driver: startInGear, and fixs.driver.set()

CARLA's automatic gearbox leaves a spawned car in neutral below about 15% throttle. Measured on vehicle.tesla.model3: 10% never moves (gear stays 0), 15% moves after 3.6 s, 20% at once; with gear 1 forced, 2% throttle moves it. A pedal-driven ego asked to creep therefore never moves.

• startInGear=True: on its first controlled step fixs.driver calls carla.ego.apply_control(carla.VehicleControl(manual_gear_shift=True, gear=1)) and writes no pedal command, so the host does not overwrite it. The autobox keeps first gear from then on, through stops. No creep at zero throttle. Off by default.
• fixs.driver.set(**options): options for every driver built after the call, so a controller can set one conditionally and keep Driver = fixs.driver(exchange) plain. An option passed to fixs.driver(...) still wins; unknown names are refused.

Kept from the first commit

LocalLink rounds values through the same float32 packet the wire carries, so an inprocess run and a run over the wire see the same numbers. Behaviour change: inprocess results shift slightly against runs made before this branch.

Related Issues

No tracker. Found while preparing the ORNL rig test (comma.ai openpilot drives the vehicle on their hub dyno). The application side is ORNL-Real-Sim/FIXS_Applications#84 (MLK eco-driving rung 6), which depends on this PR.

How to test

pytest tests/Python/unit -q

New: test_fixs_config.py; the hold (test_xil_dyno_hookup.py: off by default, a flat 0.15 m/s with aref 0 stays at rest, aref 1.5 lets go, no hold above 0.3 m/s); startInGear and fixs.driver.set (test_fixs_driver.py). End-to-end runs are on #84.

Checklist

• Code compiles/runs as expected
• Tests pass locally (tests/Python/unit: 492 passed, 8 skipped)
• Documentation is updated (doc/ConfigSetup.md: Transport row, fixs.config.get, fixs.xil.dynosim())
• Issue linked above

Not changed: the C++ bridge (VirCarlaEnv); startInGear lives in fixs.driver, so it needs nothing there.